语言: eu
数据集:

• common_voice
  评估指标:
• wer
  标签:
• audio
• automatic-speech-recognition
• speech
• xlsr-fine-tuning-week
  许可证: apache-2.0
  模型索引:
• 名称: XLSR Wav2Vec2 Large 53 Basque by pcuenq
  结果:
  • 任务:
    名称: 语音识别
    类型: automatic-speech-recognition
    数据集:
    名称: Common Voice eu
    类型: common_voice
    参数: eu
    评估指标:
    • 名称: 测试WER
      类型: wer
      值: 15.34

Wav2Vec2-Large-XLSR-53-EU

基于facebook/wav2vec2-large-xlsr-53模型，使用Common Voice数据集对巴斯克语进行微调。
使用此模型时，请确保语音输入采样率为16kHz。

使用方法

该模型可直接使用（无需语言模型），如下所示：

import torch  
import torchaudio  
from datasets import load_dataset  
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor  

test_dataset = load_dataset("common_voice", "eu", split="test[:2%]")  

processor = Wav2Vec2Processor.from_pretrained("pcuenq/wav2vec2-large-xlsr-53-eu")  
model = Wav2Vec2ForCTC.from_pretrained("pcuenq/wav2vec2-large-xlsr-53-eu")  

resampler = torchaudio.transforms.Resample(48_000, 16_000)  

# 数据预处理  
# 需要将音频文件读取为数组  
def speech_file_to_array_fn(batch):  
    speech_array, sampling_rate = torchaudio.load(batch["path"])  
    batch["speech"] = resampler(speech_array).squeeze().numpy()  
    return batch  

test_dataset = test_dataset.map(speech_file_to_array_fn)  
inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)  

with torch.no_grad():  
    logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits  

predicted_ids = torch.argmax(logits, dim=-1)  

print("预测结果:", processor.batch_decode(predicted_ids))  
print("参考文本:", test_dataset["sentence"][:2])  

评估

该模型可按以下方式在Common Voice的巴斯克语测试数据上进行评估：

import torch  
import torchaudio  
from datasets import load_dataset, load_metric  
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor  
import re  

test_dataset = load_dataset("common_voice", "eu", split="test")  
wer = load_metric("wer")  

model_name = "pcuenq/wav2vec2-large-xlsr-53-eu"  

processor = Wav2Vec2Processor.from_pretrained(model_name)  
model = Wav2Vec2ForCTC.from_pretrained(model_name)  
model.to("cuda")  

## 文本预处理  

chars_to_ignore_regex = '[\,\¿\?\.\¡\!\-\;\:\"\“\%\‘\”\\…\’\ː\'\‹\›\`\´\®\—\→]'  
chars_to_ignore_pattern = re.compile(chars_to_ignore_regex)  

def remove_special_characters(batch):  
    batch["sentence"] = chars_to_ignore_pattern.sub('', batch["sentence"]).lower() + " "  
    return batch  

## 音频预处理  

import librosa  
def speech_file_to_array_fn(batch):  
    speech_array, sample_rate = torchaudio.load(batch["path"])  
    batch["speech"] = librosa.resample(speech_array.squeeze().numpy(), sample_rate, 16_000)  
    return batch  

# 文本转换和音频重采样  
def cv_prepare(batch):  
    batch = remove_special_characters(batch)  
    batch = speech_file_to_array_fn(batch)  
    return batch  

# CPU数量或None  
num_proc = 16  
test_dataset = test_dataset.map(cv_prepare, remove_columns=['path'], num_proc=num_proc)  

def evaluate(batch):  
    inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)  

    with torch.no_grad():  
        logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits  

    pred_ids = torch.argmax(logits, dim=-1)  
    batch["pred_strings"] = processor.batch_decode(pred_ids)  
    return batch  

result = test_dataset.map(evaluate, batched=True, batch_size=8)  

# WER指标计算  
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))  

测试结果: 15.34%

训练

训练使用了Common Voice的train和validation数据集。训练共进行了22 + 20个周期，参数如下：

• 批量大小16，梯度累积步数2
• 学习率: 2.5e-4
• 激活丢弃率: 0.05
• 注意力丢弃率: 0.1
• 隐藏层丢弃率: 0.05
• 特征投影丢弃率: 0.05
• 掩码时间概率: 0.08
• 层丢弃率: 0.05